A query can specify a table name without a database or schema name (e.g.,
SELECT * FROM orders). How does CockroachDB know which
orders table is being considered and in which schema?
This page details how CockroachDB performs name resolution to answer this question.
A CockroachDB cluster can store multiple databases, and each database can store multiple tables/views/sequences. This two-level structure for stored data is commonly called the "logical schema" in relational database management systems.
Meanwhile, CockroachDB aims to provide compatibility with PostgreSQL client applications and thus supports PostgreSQL's semantics for SQL queries. To achieve this, CockroachDB supports a three-level structure for names. This is called the "naming hierarchy".
In the naming hierarchy, the path to a stored object has three components:
- database name (also called "catalog")
- schema name
- object name
The schema name for all stored objects in any given database is always
public. There is only a single schema available for stored objects because CockroachDB only supports a two-level storage structure.
In addition to
public, CockroachDB also supports a fixed set of virtual schemas, available in every database, that provide ancillary, non-stored data to client applications. For example,
information_schema is provided for compatibility with the SQL standard.
The list of all databases can be obtained with
SHOW DATABASES. The list of all schemas for a given database can be obtained with
SHOW SCHEMAS. The list of all objects for a given schema can be obtained with other
Name resolution occurs separately to look up existing objects and to decide the full name of a new object.
The rules to look up an existing object are as follows:
- If the name already fully specifies the database and schema, use that information.
- If the name has a single component prefix, try to find a schema with the prefix name in the current database. If that fails, try to find the object in the
publicschema of a database with the prefix name.
- If the name has no prefix, use the search path with the current database.
Similarly, the rules to decide the full name of a new object are as follows:
- If the name already fully specifies the database and schema, use that.
- If the name has a single component prefix, try to find a schema with that name. If no such schema exists, use the
publicschema in the database with the prefix name.
- If the name has no prefix, use the current schema in the current database.
The current database is used when a name is unqualified or has only one component prefix. It is the current value of the
database session variable.
You can inspect the list of valid database names that can be specified in
- For client apps that connect to CockroachDB using a URL of the form
postgres://..., the initial value of the
databasesession variable can be set using the path component of the URL. For example,
mydbwhen the connection is established.
The search path is used when a name is unqualified (has no prefix). It lists the schemas where objects are looked up. Its first element is also the current schema where new objects are created.
You can inspect the list of valid schemas that can be listed in
- By default, the search path contains
pg_catalog. For compatibility with PostgreSQL,
pg_catalogis forced to be present in
search_pathat all times, even when not specified with
The current schema is used as target schema when creating a new object if the name is unqualified (has no prefix).
The current schema is always the first value of
search_path, for compatibility with PostgreSQL.
- You can inspect the current schema using the special built-in function/identifier
- Index names are resolved relative to a table name using the
DROP INDEX tbl@idx;. This is the default and most common syntax.
- Index names are resolved by searching all tables in the current schema to find a table with an index named
DROP INDEX idx;or (with optional schema prefix)
DROP INDEX public.idx;. This syntax is necessary for Postgres compatibility because Postgres index names live in the schema namespace such that e.g.,
public.idxwill resolve to the index
idxof some table in the public schema. This capability is used by some ORMs.
The name resolution algorithm for index names supports both partial and complete qualification, using the same name resolution rules as other objects.
The examples below use the following logical schema as a starting point:
> CREATE DATABASE mydb;
> CREATE TABLE mydb.mytable(x INT);
> SET database = mydb;
An unqualified name is a name with no prefix, that is, a simple identifier.
> SELECT * FROM mytable;
This uses the search path over the current database. The search path is
public by default, in the current database. The resolved name is
> SET database = system;
> SELECT * FROM mytable;
pq: relation "mytable" does not exist
This uses the search path over the current database, which is now
system. No schema in the search path contain table
mytable, so the look up fails with an error.
A fully qualified name is a name with two prefix components, that is, three identifiers separated by periods.
> SELECT * FROM mydb.public.mytable;
Both the database and schema components are specified. The lookup succeeds if and only if the object exists at that specific location.
A partially qualified name is a name with one prefix component, that is, two identifiers separated by a period. When a name is partially qualified, CockroachDB will try to use the prefix as a schema name first; and if that fails, use it as a database name.
> SELECT * FROM public.mytable;
This looks up
mytable in the
public schema of the current database. If the current database is
mydb, the lookup succeeds.
For compatibility with CockroachDB 1.x, and to ease development in multi-database scenarios, CockroachDB also allows queries to specify a database name in a partially qualified name. For example:
> SELECT * FROM mydb.mytable;
In that case, CockroachDB will first attempt to find a schema called
mydb in the current database. When no such schema exists (which is the case with the starting point in this section), it then tries to find a database called
mydb and uses the
public schema in that. In this example, this rule applies and the fully resolved name is
Suppose that a client frequently accesses a stored table as well as a virtual table in the Information Schema. Because
information_schema is not in the search path by default, all queries that need to access it must mention it explicitly.
> SELECT * FROM mydb.information_schema.schemata; -- valid
> SELECT * FROM information_schema.schemata; -- valid; uses mydb implicitly
> SELECT * FROM schemata; -- invalid; information_schema not in search_path
For clients that use
information_schema often, you can add it to the search path to simplify queries. For example:
> SET search_path = public, information_schema;
> SELECT * FROM schemata; -- now valid, uses search_path
When resolving a partially qualified name with just one component prefix, CockroachDB will look up a schema with the given prefix name first, and only look up a database with that name if the schema lookup fails. This matters in the (likely uncommon) case where you wish your database to be called
> CREATE DATABASE public;
> SET database = mydb;
> CREATE TABLE public.mypublictable (x INT);
CREATE TABLE statement in this example uses a partially qualified name. Because the
public prefix designates a valid schema in the current database, the full name of
mydb.public.mypublictable. The table is created in database
To create the table in database
public, one would instead use a fully qualified name, as follows:
> CREATE DATABASE public;
> CREATE TABLE public.public.mypublictable (x INT);